Cyber-security attacks have become a disrupting issue for modern applications. In the context of autonomous vehicles, this problem can affect the users in many ways, ranging from security data breaches to a crash in the car’s system, which prevents the broad availability of these services to society. In recent years, a vast majority of studies have proposed the use of contemporary Machine Learning techniques to assist in the detection and prevention of these attacks. Nevertheless, they still can not keep up with the fast-paced nature of these adversarial attacks. In this work, we are proposing to study a trendy method called Graph Neural Networks to evaluate its efficiency and robustness over these challenges, specifically in the self-driving vehicles environment, using a publicly available dataset. The preliminary results showed that the measured metrics achieved a great performance, which paves the way for future works looking to assess stronger variables and sophisticated scenarios.